Setup for Challenge
Prerequisite
Prerequisite. Make sure you are using an RTX compatible GPU. We recommend the OS version Ubuntu 18.04/20.04 and NVIDIA driver version 525.60.11. Docker-based setup is not guaranteed to work on other OS or driver versions. We also verified the following driver versions are compatible with isaac sim:
470.141.03 535.113.01
Generate NVIDIA NGC API Key
Log in NVIDIA NGC. If you do not have an account, register one and log in.
Generate your NGC API key. You can refer to Generating API key.
Log into the NGC account on the instance
docker login nvcr.io
Type
$oauthtoken
forUsername
. Then paste your API key forPassword
. You should seeLogin Succeeded
.Make sure NVIDIA container is properly installed. Check Installation guide.
Docker Setup
Download codebase:
git clone
from thechallenge
branch.
git clone -b challenge https://github.com/arnold-benchmark/arnold.git
Move to
workspace
and build docker image.
cd arnold/workspace
docker build -f Dockerfile -t "arnold" .
sudo apt install vagrant
vagrant up
vagrant ssh
Download Data and Assets
You can download data and assets from web browser or CLI. Unzip the downloaded zip
files at /vagrant
folder.
data_for_challenge_train.zip
for trainingdata_for_challenge_val.zip
for thedev
phasedata_for_challenge_final.zip
for thetest
phasematerials.zip
for scene and object materialssample.zip
for assets used in ARNOLD
# for example, download from CLI
pip install gdown
cd /vagrant
# data_for_challenge_train.zip
gdown https://drive.google.com/uc?id=1gxSW3fFhGghJUpf_jiAy3iR_zrPy_MlJ
# data_for_challenge_val.zip
gdown https://drive.google.com/uc?id=1diLNQQcOGKEVkgOstkRbagbn_cVCKVIE
# data_for_challenge_final.zip
gdown https://drive.google.com/uc?id=1XKRxsByOwI_pYh09LUQ5wGLgzeGmL-KH
# materials.zip
gdown https://drive.google.com/uc?id=1CAT6pZfX0HqHKXU5qBdLeRZl_iY_XfOt
# sample.zip
gdown https://drive.google.com/uc?id=1jscZWcibfVXItbY1xZxRogA6Q8j3U60C
unzip data_for_challenge_train.zip
unzip data_for_challenge_val.zip
unzip data_for_challenge_final.zip
unzip materials.zip
unzip sample.zip
Model Evaluation
Please read these two evaluation scripts carefully to understand what happened:
cd /root/arnold/
bash eval_challenge_dev.sh ${checkpoint_file} # for dev phase
bash eval_challenge.sh ${checkpoint_file} # for test phase
To reproduce the results using the baseline checkpoint:
gdown https://drive.google.com/uc?id=1YuADlTFJZQc3AefULmzhZ9PrCWVjsEU2
bash eval_challenge_dev.sh peract_multi_clip_best.pth
bash eval_challenge.sh peract_multi_clip_best.pth
The first time starting the Isaac Sim takes long time, approximately 6 minutes (varying to the device configuration).
After evaluation is done, the scripts will generate an output folder. Zip the output folder (/root/arnold/output
or /root/arnold/output_dev
, depending on the phase) and submit it to EvalAI
with CLI.
zip -r submission_dev.zip -r /root/arnold/output_dev/
zip -r submission_test.zip -r /root/arnold/output/
evalai challenge 2266 phase 4500 submit --file submission_dev.zip --large --public
evalai challenge 2266 phase 4501 submit --file submission_test.zip --large --public